This paper addresses the enhancement of human-computer interaction through the synthesis of nonverbal expressions. As a means of improving the naturalness of synthesized speech for the realization of human-computer discourse, study has been focused on the effects of emotion and intention on prosodic features of human speech. Experiments were carried out to observe the time-variant behavior of the fundamental frequency, amplitude, and formant frequencies of nonverbal expressions relative to the intentions and emotional states of the speakers. From the observed data, fundamental relationships were determined and stated in the form of constraints on parameters for a formant synthesizer. Fuzzy inference was used to process these constraints. The method supports the linguistic declaration of constraints in the form of conditionals, and it enables fine-tuning through the adjustment of fuzzy set membership functions.